WASHINGTON, D.C.--Rather than try to count each and every American for the 2000 census, the U.S. Census Bureau should use sampling techniques to estimate the number of people not tallied by traditional surveys, according to a report released here today by the National Research Council (NRC). Although the census bureau still must refine its techniques to achieve a representative sample, a complete national head count--for which some members of Congress have lobbied in recent days--would be too expensive and inaccurate, the report says.
In recent censuses, the bureau has sent questionnaires by mail to every known U.S. household asking for number, gender, ethnicity, and age of all residents. Census workers have gone door-to-door afterward to collect information from unresponsive households. But this labor costs a bundle without achieving a high degree of accuracy: The 1990 census, for instance, cost $4 billion and still missed roughly 2% of Americans, the report notes.
Last year, the Census Bureau proposed that door-to-door surveys be conducted for 1% of all U.S. households as part of the 2000 census. The survey results would be used to confirm the accuracy of mail-in questionnaires and to estimate data for the uncounted population. The bureau estimates that such techniques would save about $700 million. Many Republican members of Congress have objected to such sampling, however, in part because they claim it might boost counts for urban areas, resulting in some congressional districts being redrawn in favor of Democrats.
The NRC report, from a panel chaired by statistician Keith Rust of Westat Inc., in Rockville, Maryland, backs the bureau's proposal to adopt sampling. Sampling would result in a much more accurate national head count at a more reasonable cost, says Rust. "You can afford to hire better trained people, you can get done faster so you'll be closer to the day of the census, and you can use more intensive procedures," he says. The panel praised a mock census, which included mail questionnaires and sampling visits, that the bureau carried out in three states in 1995. "Both the operations were carried out successfully, and they were able to demonstrate that they could be used effectively together," says Rust.
But the panel also tagged several areas for further improvement. For instance, survey workers should sample at least seven to 10 households in a 30-household block--as opposed to just three, as the bureau proposes. In addition, the panel recommends that the census bureau implement better procedures to ensure that sampling captures the ethnic diversity within a 30-household block.